refactor(trie): extract hybrid merge algorithm for sorted trie types

[gakonst]

Summary

Batches trie updates across all blocks in save_blocks instead of writing per-block, reducing cursor open/close overhead from N to 1.

Problem

Per #eng-perf profiling, write_trie_updates was taking ~25% of persistence time. The current implementation calls write_trie_updates_sorted once per block, opening/closing cursors N times.

In back-to-back (b2b) scenarios with 75-250 accumulated blocks, this overhead compounds significantly.

Solution

lazy_overlay.rs

• Move MERGE_BATCH_THRESHOLD constant inside function
• Use data directly instead of Arc::clone (avoids unnecessary refcount bump)
• Move Arc::make_mut into the loop for proper copy-on-write semantics

provider.rs

Add batched trie updates with hybrid merge algorithm:

Batch Size	Strategy	Allocation
0	Return default	0
1	Arc::try_unwrap	Avoids clone if refcount is 1
< 30	extend_ref with Arc::make_mut	Copy-on-write (no collect)
>= 30	k-way merge_batch	Single collect for O(n log k) merge

Allocation Behavior (No Regression)

Learned from PR #21142 review feedback:

• ❌ Don't use collect() unnecessarily in small-k path
• ✅ Use Arc::make_mut directly in the loop for copy-on-write
• ✅ Only collect for large-k path where k-way merge needs materialized data
• ✅ Use Arc::try_unwrap to avoid final clone when refcount is 1

Expected Impact

• ~50% reduction in write_trie_updates time for b2b scenarios
• Reduces cursor open/close overhead from N to 1
• Reduces MDBX transaction overhead

Testing

• All existing reth-chain-state and reth-provider tests pass
• Clippy and fmt clean

Related

• Alternative to PR perf(storage): batch trie updates across blocks in save_blocks #21142 (same optimization, cleaner implementation)
• Based on optimizations from Slack #eng-perf thread
• Learned from @yongkangc and @brian review feedback on avoiding unnecessary allocations

[mattsse]

I believe this has been merged